Researchers have developed a genius system that has the ability to oppose the lifestyle at this time. But it didn't think only a few steps – it thought that millions went forward.
The team led by the SERGEI GUKOV Mathematician from the California Institute of Technology (Caltech) has created a new learning machinery that is designed to solve mathematical problems that need a very long step like Really? Lung series or procedures; We are talking about millions.
Especially AI can proceed with the complex problems called Andrews – Korkis's guessWhich has a math person General guessing is to ask: some math puzzles can be fixed by using a set of allowances such as new arrangement or cancellation procedures?
For this reason, Caltech's new program is trying to “search for the long -term and rare steps.” Ali Shehper, the first author of the study and mathematician at Rutgers University, SARS in Caltech. statement– “It's like trying to find a way to go through the world size. These routes are a very long route that you have to test and have only one route. “
In the preprint studied post when Arxiv Last August and updated on Tuesday, Shehper and his colleagues give details that they use a newly developed AI to solve family problems related to the guess of Andrews – Curtis. Related to the abstract For clarity, they do not solve the problem of guess. While it may seem Anticlimactic Researchers have proven to refute samples that may occur continuously. While refusing the counter, it is not necessary to make the original predictions come true, but it supports it.
“The consideration of some of the samples that ensures us in the accuracy of the traditional guess and helping to create our instincts about the main problem,” Shehper explained. “It gives us a new way to think about this.” Gukov compare. Mathematics with Rubik cubic
“Can you use the cubic cubic cubic that is complicated and complex and will be taken back to the original status? You have to test these very long movements and you will not know that you are on the right path until the end, “he explained.
How did AI do? Basically, thinking outside the box According to the learning guidelines, reinforcement, researchers have trained AI by feeding for the first time, it is a math problem that is easier, followed by more difficult work. “It tries to move a variety of movements and receives a reward for solving problems. “Shehper said,” We encourage the program to do the same thing while still maintaining curiosity. In the end, it develops a new strategy that is better than humans can do. That is the magic of learning to strengthen. “
The algorithm finally learned to create an unexpected movement sequence, which researchers call “Great movement”. On the other hand, Chatgpt's results are more boring.
“If you ask Chatgpt to write a letter, it will happen to something in general. It is unlikely to happen to what is unique and original. It's a good parrot, “Gukov said.” Our program is good at finding abnormal values. “
I can think of at least one event that is convenient for AI to predict: financial problems. But while the current machine learning program is not successful in the complex level of this prognosis, the researchers predict that their methods on one day may lead to genius forecasting. that
“In general, our program knows how to learn to learn,” Gukov explains “It is thinking outside the box”. He added that the team has improved. “Mathematics updates that are decades”. Moreover, Gukov and his colleagues have given importance to the way that does not need a lot of calculations, their work can be accessed by academics. Other with small computers
Although the actual use of this success may not appear in our daily lives. But their work gathered into a host of other researchers in adjusting the learning algorithm to solve humanity. (Not to destroy our civilization)